Reproduce “Enriched Mixtures of Gaussian Process Experts” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Charles W.L. Gadd, Sara Wade, Alexis Boukouvalas

Enriched Mixtures of Generalized Gaussian Process Experts (EDP MoE) offers a scalable and interpretable approach to density regression by employing a nested clustering scheme and factorized exponential family input models. The model demonstrates competitive predictive accuracy and provides tighter, well-calibrated credible intervals on both synthetic and real-world Alzheimer's datasets, outperforming traditional GP-MoE methods.

Key insights

  • Decoupling the clustering process through EDP allows the model to infer fewer, more interpretable primary experts based on the conditional output relationship while still capturing complex input-space heterogeneity via sub-clustering.
  • The factorized input model significantly enhances scalability for multi-dimensional data and seamlessly integrates mixed input types (continuous, categorical, binomial) into a single, coherent framework.
  • The generalized GP expert framework provides broad applicability to diverse response types, making the model versatile for various density regression tasks beyond standard continuous, Gaussian-distributed outputs.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.